Recognition Of Human Actions Through DDC Transform and Silhouettes

In this work, the problem of human activity recognition is considered. A technique for human activity recognition is video sequences using space-time descriptors is proposed based on skeletons, discrete casinos transform DCT and SVM support vector machines. Variant of this technique is also presented for human activity recognition frame by frame in real time. It is based on silhouettes and DCT for characteristics extraction and RBF artificial neural networks for classification. The two proposed techniques have given very satisfied results.